Learning to Estimate the Travel Time
Zheng Wang, Kun Fu, Jieping Ye
DiDi AI Labs, Didi Chuxing
{wangzhengzwang,fukunkunfu,yejieping}@didichuxing.com
ABSTRACT
Vehicle travel time estimation or estimated time of arrival (ETA)
is one of the most important location-based services (LBS). It is
becoming increasingly important and has been widely used as a
basic service in navigation systems and intelligent transportation
systems. This paper presents a novel machine learning solution
to predict the vehicle travel time based on oating-car data. First,
we formulate ETA as a pure spatial-temporal regression problem
based on a large set of eective features. Second, we adapt dierent
existing machine learning models to solve the regression problem.
Furthermore, we propose a Wide-Deep-Recurrent (WDR) learning
model to accurately predict the travel time along a given route at
a given departure time. We then jointly train wide linear models,
deep neural networks and recurrent neural networks together to
take full advantages of all three models. We evaluate our solution
oine with millions of historical vehicle travel data. We also deploy
the proposed solution on Didi Chuxing’s platform, which services
billions of ETA requests and benets millions of customers per day.
Our extensive evaluations show that our proposed deep learning
algorithm signicantly outperforms the state-of-the-art learning
algorithms, as well as the solutions provided by leading industry
LBS providers.
KEYWORDS
Location-based services, estimated time of arrival, wide-deep-recurrent
learning
ACM Reference Format:
Zheng Wang, Kun Fu, Jieping Ye. 2018. Learning to Estimate the Travel Time.
In KDD ’18: The 24th ACM SIGKDD International Conference on Knowledge
Discovery & Data Mining, August 19–23, 2018, London, United Kingdom. ACM,
New York, NY, USA, 9 pages. https://doi.org/10.1145/3219819.3219900
1 INTRODUCTION
In recent years, the bloom of sharing economy is changing our
life in various respects. One representative is the widely used car
sharing and online ride-hailing mobile app which redenes the
way people move. Several unicorn companies are growing rapidly,
such as Uber, Lyft and Didi Chuxing, which help people to make
ecient use of the vehicles and benet millions of people per day.
The rapid expansion of this industry eld makes the location-based
service (LBS) an increasingly important problem, as an ecient and
Permission to make digital or hard copies of all or part of this work for personal or
classroom use is granted without fee provided that copies are not made or distributed
for prot or commercial advantage and that copies bear this notice and the full citation
on the rst page. Copyrights for components of this work owned by others than the
author(s) must be honored. Abstracting with credit is permitted. To copy otherwise, or
republish, to post on servers or to redistribute to lists, requires prior specic permission
and/or a fee. Request permissions from permissions@acm.org.
KDD ’18, August 19–23, 2018, London, United Kingdom
© 2018 Copyright held by the owner/author(s). Publication rights licensed to ACM.
ACM ISBN 978-1-4503-5552-0/18/08.. .$15.00
https://doi.org/10.1145/3219819.3219900
Figure 1: Estimated time of arrival (ETA) refers to the esti-
mated travel time between a pair of origin (green pin) and
destination (red pin) along the given route (green lines). It is
one of the most important location-based services for digi-
tal maps and navigation systems. It is also one of the most
important back end services for the ride-hailing apps.
precise LBS is one of the foundations for the ride-hailing platforms
to provide high quality transportation services and enjoyable travel
experience for their customers.
In this paper, we consider the estimated time of arrival (ETA) as
the travel time estimation between a pair of origin and destination
locations, which is demonstrated in Figure 1. It is an important
location-based service for digital maps and navigation systems.
ETA nds wide applications on the ride-hailing platform, as travel
time is one of the key concerns for drivers and riders to make a deal.
Thus it is essential to make an accurate estimate of the travel time
before a trip starts. An accurate ETA will boost the eciency of the
transportation system, reducing travel cost for the users, saving en-
ergy consumption and reducing motor vehicle pollution. Therefore,
ETA has become a core component that inuences decision-making
at dierent stages of the online ride-hailing process, including route
selection, vehicle dispatch, carpooling et al.
Travel time estimation has been widely studied in geographic
information systems (GIS) [
1
,
2
,
6
,
8
,
13
,
15
,
19
,
20
,
22
]. And standard
solutions have been established in the community. The existing
solutions can be divided into two categories. The rst category is
the route-based solution, which represents the travel time using an
intuitive physical model: the overall travel time of a given route is
formulated as the summation of the travel time through each road
segment
1
and the delay time at each intersection. The travel time
estimate
ˆ
y is represented as
ˆ
y =
Õ
i
ˆ
t
i
+
Õ
j
ˆ
c
j
, (1)
1
A road segment or link is dened as a route in the road network which has no
junctions at the middle but has junctions at its both ends. It can be viewed as a building
block of the road network.
Applied Data Science Track Paper
KDD 2018, August 19-23, 2018, London, United Kingdom